Sneak peak: automatic position learning using vision
Normally with RoboForth positions are programmed using a teach pad. In
JOG mode the user moves the robot in increments over and then down onto
the object to be picked up, maybe making small adjustments in X,Y,Z and
gripper angle then clicks the learn button on screen.
ST Robotics have been working on using a camera to locate an object and send coordinates to the robot controller. This includes the correct angle for the gripper. The software then learns the position as a PLACE name. The actual coordinates of the object are not of interest. The system can return to the learned position under program control.
If the position changes it may be quickly re-learned using the same vision system.
The process of learning the position involves moving the robot with camera over the object until it is in the viewport of the vision system. Rather than use the teach pad a joystick is employed which allows for continuous motion in straight lines in X,Y and Z directions.
The picture above shows the development setup. Please see the video here:
https://www.youtube.com/watch?v=KxcdBjwoNvQ
Ultimately it will be possible to learn many different positions and have different objects.
ST Robotics have been working on using a camera to locate an object and send coordinates to the robot controller. This includes the correct angle for the gripper. The software then learns the position as a PLACE name. The actual coordinates of the object are not of interest. The system can return to the learned position under program control.
If the position changes it may be quickly re-learned using the same vision system.
The process of learning the position involves moving the robot with camera over the object until it is in the viewport of the vision system. Rather than use the teach pad a joystick is employed which allows for continuous motion in straight lines in X,Y and Z directions.
The picture above shows the development setup. Please see the video here:
https://www.youtube.com/watch?v=KxcdBjwoNvQ
Ultimately it will be possible to learn many different positions and have different objects.
